Adaptive $k$NN graph model

The article introduces an adaptive graph model for the $k$-nearest neighbors ($k$NN) algorithm that addresses the trade-off between inference speed and accuracy in large-scale applications. By combining a Hierarchical Navigable Small World (HNSW) graph with a pre-computed voting mechanism, the model offloads neighbor selection and weighting to the training phase, enabling rapid navigation and precise decision boundaries. Benchmark results show significant improvements in inference speed while maintaining classification accuracy, providing a scalable solution to the limitations of traditional $k$NN methods for practitioners in non-parametric learning.